Professor McGill University, Quebec, United States
Abstract Submission: Information on the space-time variation of daily precipitation processes is essential for the management of different water systems. However, it is difficult to obtain this information based on the available historical records due to the random behaviour of these processes. Hence, stochastic approaches have been used for modelling the variability of the precipitation process. Furthermore, in the climate change context statistical approaches have been often relied on the physically unrealistic assumption that the statistical model parameters remain the same for current and future climates. Consequently, the use of stochastic approaches should be considered as more suitable to overcome this limitation of statistical methods. However, most existing stochastic methods were developed for a single site without considering the inherent spatial dependence of the precipitation processes at different locations; this could significantly affect the accuracy of impact study results. Therefore, in the present study an improved stochastic approach was proposed for modeling daily precipitations at many sites concurrently. This approach is based on a combination of two regression models for describing precipitation occurrences and amounts and on the use of the Singular Value Decomposition (SVD) technique for modelling the stochastic components of these two models. The feasibility of the proposed approach has been assessed using the available NCEP/NCAR re-analysis data and the daily precipitation data from a network of 10 raingages in Canada. It was found that the proposed multisite stochastic modeling approach was able to reproduce accurately the observed statistical properties, space-time dependence, and intermittency of the underlying precipitation processes.
